|
|
This topic includes the following sections:
The Oracle Real Application Clusters (RAC) feature supports clustering of machines that utilize replicated Oracle database services accessing the same Oracle database. Oracle RAC provides the ability to concurrently access the same Oracle database from instances physically located on multiple Oracle server machines, and offers the ability to failover unsuccessful database instances to alternate locations.
However, specific support for Oracle RAC is required by the Transaction Monitor in order to take advantage of these replication and failover features in an XA transaction environment and to obtain optimal RAC performance. This is because Oracle RAC does not allow the same database to be accessed from multiple RAC instances within the same XA transaction.
In addition, Oracle 10gR1 requires Transaction Monitor involvement when prepared transactions failover from one RAC instance to another.
Tuxedo 10.0 provides Transaction Monitor support for Oracle RAC by allowing an administrator to specify lists of groups associated with different RAC instances. This allows Tuxedo to ensure that groups associated with different instances of the same RAC database do not participate in the same transaction. The Tuxedo Oracle RAC support feature also provides a way for Tuxedo transaction manager server (TMS) processes to be notified of RAC failover events which is required when using Oracle 10gR1.
Consequently, this allows the TMS to re-obtain a list of Oracle 10gR1 prepared transactions from Oracle as required for RAC failover recovery.
When using Oracle 10gR2, administrators should use an Oracle <b>DTP Service</b> to access the Oracle RAC system, and this DTP service name should be specified in the OPENINFO string for the associated Tuxedo groups. Oracle will ensure that , and will take care of migrating the service an alternate instance if required.
The reason that patchset 10.2.0.2 or above is required for Oracle 10gR2 is due to the bug described at https://metalink.oracle.com/metalink/plsql/f?p=130:14:3193163745563425327::::p14_database_id
For specific platform software requirements, refer to the BEA Tuxedo 10.0 Platform Data Sheets in Installing the BEA Tuxedo System.
Tuxedo support for Oracle RAC requires two steps:
The following command and environment variables are used to exclusively configure Tuxedo for Oracle RAC support:
(required for all releases of Oracle) (required only for Oracle 10gR1)XARETRYINTERVAL (required only for Oracle 10gR1)Oracle 10gR1 does not allow the same database to be accessed from multiple RAC instances within the same XA transaction. In addition, Oracle 10gR1 requires Transaction Monitor involvement when prepared transactions failover from one RAC instance to another.
Oracle 10gR2 permits different RAC instances to operate on different transaction branches in RAC, but if transaction branches are on different instances, then they are loosely coupled and do not share locks. Also, for optimum commit performance, it is important to use only a single RAC instance within a given XA transaction.
For this reason, it is still important to associate an XA transaction with a single RAC instance in Oracle 10gR2. (For further information on using Oracle XA with RAC, refer to the "Developing Applications with Oracle XA" chapter in the Oracle Database Application Developer's Guide - Fundamentals.
The TUXRACGROUPS environment variable is used to associate Tuxedo groups with specific instances of Oracle RAC configurations so that Tuxedo does not include groups from multiple instances of the same RAC configuration within the same XA transaction.
A single transaction should not span multiple Oracle RAC instances. The groups that participate in a particular transaction are determined at the time the transaction is started. Each transaction is assigned to one particular instance of each RAC configuration such that the groups in each instance of a particular RAC configuration are assigned to an equal number of transactions.
TUXRACGROUPS
TUXRACGROUPS environment variable specifies the groups that are associated with a particular RAC configuration, and will disallow sending service calls in the same transaction to two or more groups identified as different instances of the same RAC configuration.
| WARNING: | The TUXRACGROUPS environment variable must be set on all machines in a configuration, and must have the same sets of groups specified in the same order on all machines. |
| WARNING: | If this restriction is not followed, then inconsistent sets of groups can be included within a transaction. The coordinating group will notice the inconsistency at commit time, roll back the transaction, and send an error message to the userlog. |
The TUXRACGROUPS environment variable is used to define Oracle RAC group configurations. Its syntax is as follows:
TUXRACGROUPS="G1,G2,...,Gm;H1,H2,...,Hn[;...]:I1,I2,...,Io;J1,J2,...,Jp[;...][:...]"
| Note: | Typically, most users place all of the services associated with one database instance in a single group, therefore commas are not needed in the TUXRACGROUPS value. |
TUXRACGROUPS environment variable is to specify groups associated with different instances of the same Oracle RAC configuration, all applications using the TUXRACGROUPS variable should have at least one semicolon in the environment variable value.
| Note: | Typically, most users specify only one RAC database configuration, therefore colons are not needed in the TUXRACGROUPS value. |
This section describes four different examples for defining Oracle RAC group configurations:
Figure 6-1 shows a simple Oracle RAC configuration.
In this example, there is one Oracle database, (ORA1), two Oracle RAC instances with 1 group per each instance.
The same transaction request to both GROUP1 and GROUP2 cannot be sent because they access database services through different instances that map to the same Oracle RAC database configuration.
TUXRACGROUPS="GROUP1;GROUP2:GROUP3;GROUP4,GROUP5"
Figure 6-2 shows an example of adding multiple groups to a single instance.
In this example, there are two Oracle databases: ORA1 and ORA2. ORA1 offers machine-specific services ORA1SITE1 and ORA1SITE2, and ORA2 offers machine-specific services ORA2SITE1 and ORA2SITE2. The objective is to assign an approximately equal number of transactions and configure the same services to the groups associated with each instance of an Oracle RAC configuration.
The same transaction request to both GROUP1 and GROUP2 cannot be sent because they access database services through different instances that map to the same Oracle RAC database configuration. The same applies to GROUP3 and GROUP4 or GROUP3 GROUP5, the same transaction cannot be sent to both these groups.
GROUP4 and GROUP5 both access the same database service of the same Oracle RAC database configuration, so these groups would be permitted together. GROUP1 and GROUP4 would be permitted together, because they access different RAC database configurations. If there is also a GROUP6 in this configuration, it would be permitted with any other group, because GROUP6 is not an Oracle RAC group.
| Note: | The number of groups in each Oracle RAC instance does not have to be the same. |
The *GROUPS and *SERVERS sections of the UBBCONFIG file for this configuration might look as follows:
*GROUPS
DEFAULT: TMSNAME=TMS_ORA TMSCOUNT=2
GROUP1 LMID=SITE1 GRPNO=1
OPENINFO="ORACLE_XA:Oracle_XA+Acc=P/scott/tiger+SqlNet=ORA1SITE1+SesTm=100+LogDir=.+MaxCur=5"
GROUP2 LMID=SITE2 GRPNO=2
OPENINFO="ORACLE_XA:Oracle_XA+Acc=P/scott/tiger+SqlNet=ORA1SITE2+SesTm=100+LogDir=.+MaxCur=5"
GROUP3 LMID=SITE1 GRPNO=3
OPENINFO="ORACLE_XA:Oracle_XA+Acc=P/scott/tiger+SqlNet=ORA2SITE1+SesTm=100+LogDir=.+MaxCur=5"
GROUP4 LMID=SITE2 GRPNO=4
OPENINFO="ORACLE_XA:Oracle_XA+Acc=P/scott/tiger+SqlNet=ORA2SITE2+SesTm=100+LogDir=.+MaxCur=5"
GROUP5 LMID=SITE2 GRPNO=5
OPENINFO="ORACLE_XA:Oracle_XA+Acc=P/scott/tiger+SqlNet=ORA2SITE2+SesTm=100+LogDir=.+MaxCur=5"
GROUP6 LMID=SITE1 GRPNO=6 TMSNAME=TMS_QM
OPENINFO="TUXEDO/QM:/home/myapplication/QUE:QSPACE"
*SERVERS
DEFAULT: RESTART=Y MAXGEN=5 REPLYQ=Y CLOPT="-A"
EMPLOYEE_SVR SRVGRP=GROUP1 SRVID=1
EMPLOYEE_SVR SRVGRP=GROUP2 SRVID=2
BANKING_SVR SRVGRP=GROUP3 SRVID=3
BANKING_SVR SRVGRP=GROUP4 SRVID=4
BANKING_SVR SRVGRP=GROUP5 SRVID=5
| Note: | GROUP4 and GROUP5 have the same OPENINFO strings, because they both use the same database service from the same database. |
| Note: | The specification of the OPENINFO string for Oracle groups in the *GROUPS section is the same as when using Oracle without RAC. For information on how to specify an OPENINFO string for an Oracle group, refer to the Developing Applications with Oracle XA chapter in the Oracle Database Application Developer's Guide - Fundamentals. |
TUXRACGROUPS="GROUP11,GROUP12,GROUP13;GROUP21,GROUP22:GROUP3;GROUP4, GROUP5"
Figure 6-3 shows an example of adding multiple groups to multiple instances.
This example is similar to the previous example — except that GROUP11, GROUP12, and GROUP13 are all associated with the first RAC instance of the first RAC configuration, and GROUP21 and GROUP22 are both associated with the second RAC instance.
If the first service call in a transaction in this configuration goes to GROUP12, then it would be possible to send other service calls in this transaction to GROUP11, GROUP12, or GROUP13, but not to GROUP21 or GROUP22.
If a transactional service call is made to a service that is not advertised in any permitted groups but is available in one or more prohibited groups, the result is:
For each RAC configuration defined as part of the TUXRACGROUPS environment variable, Tuxedo determines which RAC group(s) in that configuration participate in a particular transaction when that transaction is started.
TUXRACGROUPS="GROUP1A,GROUP2A;GROUP1B,GROUP2B"
Figure 6-4 shows an example of routing transactional and non-transactional requests in an Oracle RAC configuration.
GROUP1A and GROUP2A are in RAC instance 1 and GROUP1B and GROUP2B are in RAC instance 2.
Data dependent routing for transactional services offered in RAC groups is likely to achieve the desired result only if:
Since all but one of the instances in a RAC configuration are disallowed in a particular transaction, each data value must be specified for a service in each RAC instance. Otherwise, that data value will not be processed by any service in the RAC configuration for some transactions.
If all data values are processed by the same set of service instances, then there is no need to use data dependent routing.
When routing occurs, any groups that are not permitted for the current transaction are ignored. The routing decision only considers:
If routing is performed for a non-transactional request, all groups can participate. The service is routed to the first group matching the data value listed in the UBBCONFIG file *ROUTING section RANGES field.
If routing is performed for a mixture of transactional and non-transactional requests, some applications may not require non-transaction request load balancing. You can vary the RAC instances listed first in your application for different data values so that non-transactional requests are balanced accordingly among services offered by different RAC instances.
The *SERVICES and *ROUTING sections of the UBBCONFIG file for this configuration might look as follows:
*SERVICES
DEPOSIT SRVGRP=GROUP1A ROUTING=MYROUTE
DEPOSIT SRVGRP=GROUP2A ROUTING=MYROUTE
DEPOSIT SRVGRP=GROUP1B ROUTING=MYROUTE
DEPOSIT SRVGRP=GROUP2B ROUTING=MYROUTE
*ROUTING
MYROUTE FIELD="BRANCH_ID"
RANGES="1-5:GROUP1A, 1-5:GROUP1B, 6-10:GROUP2B, 6-10:GROUP2A, *:*"
BUFTYPE="FML32"
Requests with a BRANCH_ID of 1 through 5 must be handled by GROUP1A or GROUP1B. Requests with a BRANCH_ID of 6 through 10 must be handled by GROUP2A or GROUP2B.
| Note: | The RANGES value should duplicate each routing data value. It should specify a service associated with each RAC configuration instance in the RANGES field. |
For transactional requests, the first half, branches 1-5 map to GROUP1A and branches 6-10 map to GROUP2A; the other half branches 1-5 map to GROUP1B and branches 6-10 map to GROUP2B.
For non-transactional requests, branches 1-5 map to GROUP1A, and branches 6-10 map to GROUP2B because these are the first groups specified that match the respective routing ranges.
Requests with an invalid BRANCH_ID are mapped to any permitted group.
TMS_rac_refresh(1), XARETRYDURATIONSECONDS, and XARETRYINTERVAL specifically handle transaction recovery issues.
TMS_rac_refresh(1)is called when an Oracle RAC group fails over to an alternate group. TMS_rac_refresh(1) should not be executed manually from the command line; the proper way to invoke TMS_rac_refresh(1) is to use Oracle Fast Application Notification (FAN).
| Note: | For more details on configuring Oracle FAN, refer to Oracle 10g documentation. |
The XARETRYDURATIONSECONDS and XARETRYINTERVAL environment variables are used to retry transaction recovery operations (xa_recover()) as required by Oracle RAC.
XARETRYDURATIONSECONDS
xa_recover() operations when TMS_rac_refresh(1) is called. If it is not set or set to 0, then xa_recover() is performed once only.
XARETRYDURATIONSECONDS is 0.
| Note: | For Oracle 10.1, it is recommended that XARETRYDURATIONSECONDS is set to 120. |
XARETRYINTERVAL
xa_recover() operations are retried during the XARETRYDURATIONSECONDS interval. The XARETRYINTERVAL value is relevant only if XARETRYDURATIONSECONDS is set to a value greater than 0.
A key process in configuring Tuxedo for Oracle RAC is setting up Oracle FAN to invoke TMS_rac_refresh(1) with the appropriate group parameter on group failover. (More group parameter and group failover information is provided in Configuring Transaction Propagation.)
More information regarding Oracle FAN can be found in the Workload Management with Oracle Real Application Clusters (PDF) White Paper.
The following is an example of an Oracle FAN script.
//This File should be placed at ORA_CRS_HOME/racg/usrco//
------------------------------calout.sh-------------------------
#! /bin/ksh
#parse the event
AWK=awk
NOTIFY_EVENTTYPE=$1 # Event type is handled differently
for ARGS in $*
do
PROPERTY=`echo $ARGS|$AWK -F"=" '{print $1}'`
VALUE=`echo $ARGS|$AWK -F"=" '{print $2}'`
case ${PROPERTY} in
VERSION|version)NOTIFY_VERSION=$VALUE;;
SERVICE|service)NOTIFY_SERVICE=$VALUE;;
DATABASE|database)NOTIFY_DATABASE=$VALUE;;
INSTANCE|instance)NOTIFY_INSTANCE=$VALUE;;
HOST|host) NOTIFY_HOST=$VALUE ;;
STATUS|status) NOTIFY_STATUS=$VALUE;;
REASON|reason) NOTIFY_REASON=$VALUE;;
CARD|card) NOTIFY_CARDINALITY=$VALUE ;;
TIMESTAMP|timestamp) NOTIFY_LOGDATE=$VALUE;; # catch
event
??:??:??) NOTIFY_LOGTIME=$PROPERTY;; # catch event time
(hh24:mi:ss)
esac
done
#Set the REFRESH_DIR environment variable.
. /home/oracle/callout.env
#Make a log to record events.
FAN_LOGFILE=/home/oracle/app/products/10.1.0.3.0/db_1/calloutlog/`hostname`_upti
me.log
touch ${FAN_LOGFILE}
echo ${1} >>${FAN_LOGFILE}
#invoke the TMS_rac_refresh command.
if [ ${NOTIFY_EVENTTYPE} = "INSTANCE" -a ${NOTIFY_STATUS} = "down" ]
then
${REFRESH_DIR}/rac_refresh >> ${FAN_LOGFILE} 2>&1
fi
-----------------------------callout.sh end-----------------------------
-----------------------------callout.env----------------------------------
#! /bin/ksh
#TUXEDO and Oracle RAC server are not one the same machine.
export REFRESH_DIR=/tmp
-----------------------------callout.env end------------------------------
-----------------------------rac_refresh----------------------------------
#! /bin/ksh
#If TUXEDO and Oracle RAC server on different machine
. /home/oracle/callout.env
rsh -l ${LOGNAME} ${TUX_MASTER_MACHINE} ${REFRESH_DIR}/rac_refresh
>/tmp/run1.log 2>&1
rsh -l ${LOGNAME} ${TUX_NONMASTER_MACHINE}
${REFRESH_DIR}/rac_refresh >/tmp/run1.log 2>&1
#If TUXEDO and Oracle RAC server are on same machine
#set up environment variable
#export APPDIR=/tmp
#export ORACLE_HOME=/home/oracle/Ora10g
#export TUXDIR=/nfs/users/libo/r902/BJ/bld
#export PATH=.:${PATH}:${TUXDIR}/bin
#. $TUXDIR/tux.env
#export TUXCONFIG=${APPDIR} /tuxconfig
#invoke TMS_rac_refresh
#TMS_rac_refresh RACDBGRP1
#TMS_rac_refresh RACDBGRP3
-----------------------------rac_refresh end-----------------------------
For Oracle 10gR2, it is much simpler to configure transaction recovery. The database services specified in the OPENINFO string for each group associated with Oracle RAC should be declared in Oracle as DTP services.
For example, in Listing 6-1, GROUP1 accessed Oracle via service ORA1SITE1 and GROUP2 accessed Oracle via service ORA1SITE2. In Oracle 10gR2, service ORA1SITE1 should be declared with DTP=TRUE, with preferred instance SITE1, and with available instance SITE2. Service ORA1SITE2 should be declared with DTP=TRUE, with perferred instance SITE2, and with available instance SITE1. A similar process should be followed for groups GROUP3, GROUP4, and GROUP5.
By declaring different perferred instances, the application will be able to get the benefit of load balancing during normal operation when both instances are available.
The setting of the TUXRACGROUPS enviornment variable will ensure that different instances of the RAC configuration are not combinined in the same transaction in order to obtain optimal performance. If one of the RAC instances goes down, Oracle will transfer the DTP service to the non-preferred instance while maintaining transactional integrity.
When using Oracle 10gR2 DTP services, it is not necessary and is not recommended to configure Oracle FAN, use TMS_rac_refresh(1) or set the XARETRYDURATIONSECONDS or XARETRYINTERVAL environment variables.
Although the Tuxedo Oracle RAC environment variables can be initiated at the operating system command line, it is highly recommended that you use the ENVFILE parameter specified in the *MACHINES section of the UBBCONFIG file to initiate these environment variables.
Apply the following syntax considerations when setting the environment variables for Oracle RAC.
ENVFILE, which is the preferred method, quotation marks are not permitted around the environment variable value.
|
